P1 begins generation with material properties, loads, and boundary conditions, then searches the resulting shape families for configurations that satisfy the constraints. Geometry is not treated as primary while material is deferred. This ordering follows the physics: geometry is visible and intuitive, while material specification is often treated as a downstream detail.
P1 does the reverse. The P1 physics engine starts with material properties and derives admissible shape families instead of assigning material to a fixed shape. This changes the reachable design regions, the accuracy of tradeoffs, and which candidates remain in the validated output set. That is the architectural basis.
How Material Properties Limit Shape Space
Component performance couples three inputs: material, geometry, and loading. Geometry cannot be optimized without material, since its structural effect depends on elastic modulus, yield strength, density, and thermal conductivity. A topology optimal for Ti-6Al-4V at 70 GPa elastic modulus and 4.43 g/cc density is not optimal for aluminum 7075-T6 at 71 GPa elastic modulus and 2.81 g/cc density, despite nearly identical elastic moduli. The 1.62 g/cc density difference changes the mass penalty for a given second moment of area, so the minimum-mass topology at a given stiffness also changes.
This is more than a final factor-of-safety adjustment. For given loads and material, optimal topology depends on specific stiffness, specific strength, and their relation to the geometric efficiency of structural forms. Thin-walled tubes suit materials with high specific stiffness in bending. Lattice structures suit materials compatible with additive manufacturing when mass-specific stiffness under multiaxial loading is the constraint. Open-section profiles suit some axial cases but fail in torsion unless shear modulus is high relative to density.
Material-first generation gives the search algorithm the correct constraint space for the actual material from its first evaluation. Geometry-first generation with later material assignment spends computation on shape regions suboptimal for that material, reducing candidate density near the true optimum.
Constraint Definition Sequence
In P1, each session begins with the constraint definition block. The engineer enters, in order: material; load cases, with magnitudes and directions; safety factor; boundary conditions and fixture locations; and design-domain dimensions. The algorithm builds its searched shape space in that sequence, with material properties first constraining which shapes are physically sensible.
Material entry does more than label the design. It fills the stiffness matrix parameters, yield criterion threshold, density field mapping geometry to mass, and thermal conductivity tensor when thermal constraints are active. Later generation steps use these values directly. For a proposed member with a given cross-section and length, the admissibility check computes axial and bending stiffness from the specified elastic modulus, compares them with load-induced stress, and marks the member admissible or inadmissible. No shape is evaluated without material-property awareness.
Multi-Material Consequence
Material-first architecture makes multi-material generation a primary problem, not post-processing. With geometry-first generation, assigning a second material to component regions requires recalculating composite stiffness, often changing which features carry load. The optimization may need to run again, or produce a suboptimal result because the topology was not designed for its final material distribution.
With material properties first, the algorithm can determine assignment and geometry together. A candidate may place material A in primary members under axial loading, where specific stiffness dominates, and material B in thermal interface regions, where thermal conductivity matters more. The constraint check evaluates the composite under the full constraint set, rather than either material alone. This costs more computation than single-material generation, but yields candidates representing achievable multi-material performance instead of a single-material result corrected afterward.
Our initial implementation targets homogeneous single-material generation because it covers most use cases we address directly. Full physics-consistent multi-material generation is an architectural extension, not a retrofit. The constraint definition block already accepts material assignment per region rather than globally, and stiffness matrix construction supports heterogeneous distributions.
Density-Based Design Representation
Internally, the physics engine represents topology as a density field over the design domain, not an explicit boundary. Each voxel has a density between zero and one: zero is void and one is solid material. Under SIMP (Solid Isotropic Material with Penalization), voxel stiffness scales with density raised to a penalty exponent, typically 3. This penalizes intermediate values and drives solutions toward binary solid-void distributions.
Material properties enter directly: a full-density voxel has the specified material's elastic modulus. As optimization drives densities toward zero or one, high-density voxels become structural members and near-zero voxels become voids. The final geometry comes from thresholding the density field at typically 0.5, then generating a surface mesh from the binary field.
This representation couples material and generation because the FEA solver computes each iteration's stiffness tensor directly from material properties and current density. In this formulation, material and shape remain coupled at every optimization evaluation.
What Shape-First Methods Miss
Shape-first methods with post-hoc material assignment are faster to set up and produce plausible topologies across more inputs, including vague or underspecified constraints. For a concept-phase sketch of a possible component, they are appropriate.
They are unsuitable when each candidate's structural performance must be accurate for the intended material. If optimization used implicit material assumptions unlike the production material, its mass-stiffness tradeoffs are wrong. The error is systematic, with direction set by how the actual material's density-to-stiffness ratio differs from the assumption. For aerospace candidates with tight mass budgets and binding stiffness requirements, these errors make candidates appear better than they are and understate the material choice's performance cost.
This is why P1's physics engine begins with material properties. Its candidates are evaluated against the material the engineer intends to use, and reported tradeoffs reflect achievable performance. That accuracy supports reliable structural engineering judgment. Without it, judgment is applied to potentially misleading results, which is worse than having fewer honest results.